now 1557 from 1350
tbh, i dont feel my mmr changes because I check every game I queue in "live game" section in the end
its very solid around page 10-20 all of them after first 10 games, but seldom get into first page or top 5 pages
btw, I solo queue all time and cross the section not to queue with stacks
I'd suggest you make a blog for changelog of everything so you won't have to go through all the trouble of answering everybody's questions. Or maybe a FAQ section.:)
I think I've been rated too high. I feel I should only be around 800-1000 while the page rated me 1587.
thats kinda like valves mmr you lose rating while winning games thats true especially in lpq each time i leave lpq even on winstreaks its like my account got rating reset
It's nice to see you have a good rating, but after seeing lots of ultra high DSR's from really bad players, i really dont care anymore :) I just hope i'll get people of my skill in my and oposite teams when i play, and thats it :)
Cheers. I'm really interested to see the variance among matches. The skill level (at least from my perspective) changes a ridiculous amount, and the few DSRs that I've been able to see pretty much back up what I have felt. I just don't see how I can dominate matches against players at a certain level, and then suddenly queue into matches which are all 900-1000 DSR lower than the people that I just dropped..
The creator has changed the algorithm in an attempt to judge a players individual performs per game.
So basically, it doesn't take into account how difficult your opponents are, or you sacrificing yourself for the team etc it seems. Honestly i'm not sure, but the whole thing seems completely meaningless in this iteration.
I went from 900 to 200 after going on a nice win streak.
@Concede
It takes into account how difficult your opponents were and just how good you've played against them. If you stomped but your opponents were bad, you can have low DSR because it's easy thing to achieve ( and it will be low unless you've stomped them to 0 deaths or pushed almost all towers alone or something like that).
@krkshka
If you played badly or played mediocre vs bad opponents, your DSR will be low. Examples from your links:
The first 3 mason games:
419736491 - 251
He went 0:1:0 with furion ( ragequit? )
419703481 - 323
Again, he went 0:3:0 with furion in a 23 minute game.
419649143 - 665
Again, 2:7 with furion - getting carried by Lina.
Now for the better games of this same player:
419566291 - 1317
15:2 with Lifestealer
418754417 - 1956
4:8 with Storm - this game was hard for him as he had 40% of team kills and 3 of his teammates went 1:7, 1:9, 1:7 - so even if he had little kills, he still performed well in regards to his team. If you played this match in Valve's matchmaking, you would lose a lot of rating. DSR tried to compensate for bad teammates - and he got 1956 points for this.
418715610 - 1904
14:6 With slark - and he lost because of his team again. He performed well and got a lot of score for that.
As for Funnik:
419576038 - 987
19:10 as pudge - most deaths of his team. 2 other, non-mid players did as much damage as him and didn't die as often. In my opinion, this game wasn't the best.
419551921 - 1674
11:4 with clinkz and 947 tower damage - not that bad but not a 1900 level game.
419526587 - 1048
20:3:10 with PA - seems like a good KDA for this low rating. Well apparently, his opponents weren't that good in that game and he should have probably went 20:1:10 or even 20:0:10 for 1900 rating. Maybe he stacked with worse players or simply faced an opponent(s) with a bad day - because the enemies didn't seem to play top-level dota.
417057007 - 1753
14:14:27 - mediocre kda but he faced FATA- and paS in one game so he got a really nice score.
Now for his highest game:
404551639 - 1964
20:6:14 as Luna - stacked with XBOCT and Dendi ( so he got top level opponents ) and managed to get nice stats.
Again, due to performance reasons, for some people there are only 2-3 games - this number will soon increase and possibly you will be able to get a score in real time after pressing a button ( so you play a game and see how much DSR you've got for it almost instantly ).
Sometimes you have bad games and they will lower your score artifically as right now, for some accounts, it only takes into account your last 3 games.
Even top level players (Funnik, mason ) have bad-mediocre games where they lose hard or get stomped - it's just that, on average, they have way less such games than normal players and if you've looked at DSR for last 50 games - their total DSR would be way higher. Once I get more performance, it will become visible.
Well you have 2 games, it's not a significant amount to make any conclusions yet. Besides, each version is more accurate and it doesn't really matter that there are some people that saw swings of 1000 versus the previous versions - it usually means you were an edge case that's being rated more accurately now due to new signals/fixing bugs.
KDA ratios are generally an awful method to judge a player's skill, but I guess if you want to differentiate the score from Valve's MMR, it could be somewhat useful... for some people.
This calculation website is absofrakkinglutely hokus pokus pedler crap... you play with a team on a support role and stomp on the others you get crap Rating, you loose and play well you get a positive rating, the next game is negative...
You buy wards, deward crap, use mek, use pipe, pull guys to you, save teammates none of this is taken into account..
seems like the site is made to glorify something that the OP considers to be the right way to play and screw all the rest....
I'll go with the detractors of this website and completely forget it, these numbers dont make sense at all.
I am all for you continuing to work on your site and improving your algorithm, but your site continuously makes extreme changes to players ratings, and has done so at a time when an official MMR has been released.
I wish you the best of luck, but the communities skepticism is justified.
For example, why my DSR in this match http://dotabuff.com/matches/425664634 was 779? My team was terrible and I almost won the game by myself. You can see my hero damage is much higher than anybody in my team, and GPM was close to AM which is very good.
Then in this one http://dotabuff.com/matches/425741824, even though I played better than an average Tinker, my performance was not good, and we lost. My KDA wasn't high, only decent. Yet I got 1281 DSR.
I am working on something that will answer most of you criticisms :) I think it will be finally something that will really be "it".
I am sorry that the scores seem unstable - it's still a work in progress and a beta - I use your feedback to improve the whole method. So your criticism is welcome - I read a lot of what you write and try to incorporate as much interesting ideas as I can.
Everyone has bad games, even Funnik, Dendi etc. it's just that they have less of them and in the long run their DSR would be higher. Right now you make conclusions on limited data - 2,3,5 matches while this gets stable at around 20-30. You might have been just unlucky and got a bad game streak. Again, this should be resolved once I optimize the performance a bit.
As for how this currently works:
Right now this version answers two basic questions:
How hard was this game?
How well did this player play given those circumstances?
Then it converts the answer into a score.
So while you may think you've played well, the algorithm looked at your opponent stats and saw that they didn't seem to play top notch dota - that's why you got lower score.
You might have had negative kill:death ratio but the algo saw that opponents played really well ( because they started tryharding, were 5 man stack, all of them picked their best herores, hated you because you insulted them in all-chat so they started playing more intensely etc. ) and the algorithm tried to compensate for this and give you higher score.
Still, tomorrow or in 2 days there will be a new version that I think everyone will find useful, regardless of Valve's MMR :)
"So while you may think you've played well, the algorithm looked at your opponent stats and saw that they didn't seem to play top notch dota - that's why you got lower score."
It gives me >2000 points for a one sided stomp game which I ended with 4-1 score...
Its not JUST your performance. Its your performance vs "expected" performance. Which essentially is comparing both lineups to determine who should win, and this will make stomps in games where you should be winning give less points, and losing in matches which you should win less points. Add in performance, and what you get is that because the game was unbalanced in your favor, to get a higher score you need to do a lot better. A 3.1 KDA might be decent in a game which is balanced or unbalanced against you, but if its in your favor, then its not a very good score.
he explains this above
>20:3:10 with PA - seems like a good KDA for this low rating. Well apparently, his opponents weren't that good in that game and he should have probably went 20:1:10 or even 20:0:10 for 1900 rating. Maybe he stacked with worse players or simply faced an opponent(s) with a bad day - because the enemies didn't seem to play top-level dota.
@Terrible, I got 2091 points for this game: http://dotabuff.com/matches/423210041 I think the current algorithm doesn't work properly. Furthermore, I'm playing against people on the first live game pages and my matches get rated between 600 and 1200 points.
On the other side, I raped the #5 player in the list (Ming明明) in two consecutive games (although I lost the second one due to feeding team mates :) ), I think he is pretty overrated.
can someone check mine pls? been trying to get my rating T_T
http://dotabuff.com/players/112003723
thanks ^^
http://mydotaskill.com/index.php?id=114855365 1604 8-)
I lost 284 points because I increased my winrate to 75%...
http://mydotaskill.com/index.php?id=20355271
My Invoker 1/12/5
My Pudge 1/14/5
Still above 1300. =D
saigyou such lowskilled ;/
http://mydotaskill.com/index.php?id=112003723
Wow, my score lowered from 1022 to 196 :D
I guess that means my friends deter my improvement, because after getting that 1022 I've only played in 2-4 man stacks. Solo queue here I go :P
from 1646 to 1052 lal xD
1542
http://mydotaskill.com/index.php?id=http%3A%2F%2Fdotabuff.com%2Fplayers%2F103009657
P.s. What does "from" mean ? I don't understand that part.
now 1557 from 1350
tbh, i dont feel my mmr changes because I check every game I queue in "live game" section in the end
its very solid around page 10-20 all of them after first 10 games, but seldom get into first page or top 5 pages
btw, I solo queue all time and cross the section not to queue with stacks
can you calc http://dotabuff.com/players/3221920 me for some reason all i get is a "." when i do it
Could you show us roughly what scores equate to what skill tier (Normal/High/Very High)?
Soon. Please let me finish this new algorithm. Once it's rolled out, you will be able to see how close to pro level you are. :)
My score hasn't changed. And I played more than twenty games in 6.79. Any idea why?
@Fearohnum
I'd suggest you make a blog for changelog of everything so you won't have to go through all the trouble of answering everybody's questions. Or maybe a FAQ section.:)
I think I've been rated too high. I feel I should only be around 800-1000 while the page rated me 1587.
Might I have my super low score calculated please?
i lost 447 cus i won 18games from my last 20 matches...seems fair x)
thats kinda like valves mmr you lose rating while winning games thats true especially in lpq each time i leave lpq even on winstreaks its like my account got rating reset
It's nice to see you have a good rating, but after seeing lots of ultra high DSR's from really bad players, i really dont care anymore :) I just hope i'll get people of my skill in my and oposite teams when i play, and thats it :)
I have 731 Platinum...from my stand point..it work.. I guess?
The whole process and algorithm were changed. New scores are being calculated and you can see how much DSR you've got for each game.
Also, the TOP player rankings now refresh in real time. :)
Cheers. I'm really interested to see the variance among matches. The skill level (at least from my perspective) changes a ridiculous amount, and the few DSRs that I've been able to see pretty much back up what I have felt. I just don't see how I can dominate matches against players at a certain level, and then suddenly queue into matches which are all 900-1000 DSR lower than the people that I just dropped..
What does a negative DSR score in recent games mean?
My DSR is somewhere around 200
MMR around 4300
REASON NOW !!
The creator has changed the algorithm in an attempt to judge a players individual performs per game.
So basically, it doesn't take into account how difficult your opponents are, or you sacrificing yourself for the team etc it seems. Honestly i'm not sure, but the whole thing seems completely meaningless in this iteration.
I went from 900 to 200 after going on a nice win streak.
^ So that explains why one of my friends got 5 in dsr rating after the update XD
This seems to be the least reliable iteration of your algorithm yet Fearohnum. Just look at this guy: http://mydotaskill.com/index.php?id=121847953 The difference in DSR among those matches is quite insane. Here is another example: http://mydotaskill.com/index.php?id=86723143
@Concede
It takes into account how difficult your opponents were and just how good you've played against them. If you stomped but your opponents were bad, you can have low DSR because it's easy thing to achieve ( and it will be low unless you've stomped them to 0 deaths or pushed almost all towers alone or something like that).
@krkshka
If you played badly or played mediocre vs bad opponents, your DSR will be low. Examples from your links:
The first 3 mason games:
419736491 - 251
He went 0:1:0 with furion ( ragequit? )
419703481 - 323
Again, he went 0:3:0 with furion in a 23 minute game.
419649143 - 665
Again, 2:7 with furion - getting carried by Lina.
Now for the better games of this same player:
419566291 - 1317
15:2 with Lifestealer
418754417 - 1956
4:8 with Storm - this game was hard for him as he had 40% of team kills and 3 of his teammates went 1:7, 1:9, 1:7 - so even if he had little kills, he still performed well in regards to his team. If you played this match in Valve's matchmaking, you would lose a lot of rating. DSR tried to compensate for bad teammates - and he got 1956 points for this.
418715610 - 1904
14:6 With slark - and he lost because of his team again. He performed well and got a lot of score for that.
As for Funnik:
419576038 - 987
19:10 as pudge - most deaths of his team. 2 other, non-mid players did as much damage as him and didn't die as often. In my opinion, this game wasn't the best.
419551921 - 1674
11:4 with clinkz and 947 tower damage - not that bad but not a 1900 level game.
419526587 - 1048
20:3:10 with PA - seems like a good KDA for this low rating. Well apparently, his opponents weren't that good in that game and he should have probably went 20:1:10 or even 20:0:10 for 1900 rating. Maybe he stacked with worse players or simply faced an opponent(s) with a bad day - because the enemies didn't seem to play top-level dota.
417057007 - 1753
14:14:27 - mediocre kda but he faced FATA- and paS in one game so he got a really nice score.
Now for his highest game:
404551639 - 1964
20:6:14 as Luna - stacked with XBOCT and Dendi ( so he got top level opponents ) and managed to get nice stats.
Again, due to performance reasons, for some people there are only 2-3 games - this number will soon increase and possibly you will be able to get a score in real time after pressing a button ( so you play a game and see how much DSR you've got for it almost instantly ).
Sometimes you have bad games and they will lower your score artifically as right now, for some accounts, it only takes into account your last 3 games.
Even top level players (Funnik, mason ) have bad-mediocre games where they lose hard or get stomped - it's just that, on average, they have way less such games than normal players and if you've looked at DSR for last 50 games - their total DSR would be way higher. Once I get more performance, it will become visible.
this site makes 0 sense I go from 1200 to 400 to 1200 seems like you are pulling numbers out of your ass but w.e
Well you have 2 games, it's not a significant amount to make any conclusions yet. Besides, each version is more accurate and it doesn't really matter that there are some people that saw swings of 1000 versus the previous versions - it usually means you were an edge case that's being rated more accurately now due to new signals/fixing bugs.
^
srsly how the hell is my rating that low?
Your algorithm has bugs, some games are rated with dsr < 0.
calculate mine please <3
KDA ratios are generally an awful method to judge a player's skill, but I guess if you want to differentiate the score from Valve's MMR, it could be somewhat useful... for some people.
Yeah because KDA was the only factor he used.
I have 15 games in a row that alternate between 776 DSR and 923.
Forget DSR, play Dota 2 as a teamgame, enjoy.
That method is fucked up hard. Play a good game: >600 score; play like you never saw dota before: <1400 score...
there is nothing accurate about this.
play a good game - 90000000 score, play a bad game - 00000001 score. website is useless
This calculation website is absofrakkinglutely hokus pokus pedler crap... you play with a team on a support role and stomp on the others you get crap Rating, you loose and play well you get a positive rating, the next game is negative...
You buy wards, deward crap, use mek, use pipe, pull guys to you, save teammates none of this is taken into account..
seems like the site is made to glorify something that the OP considers to be the right way to play and screw all the rest....
I'll go with the detractors of this website and completely forget it, these numbers dont make sense at all.
@Fearohnum // MyDotaSkill.com
I am all for you continuing to work on your site and improving your algorithm, but your site continuously makes extreme changes to players ratings, and has done so at a time when an official MMR has been released.
I wish you the best of luck, but the communities skepticism is justified.
Fearohnum, your algorithm seems a bit confusing
For example, why my DSR in this match http://dotabuff.com/matches/425664634 was 779? My team was terrible and I almost won the game by myself. You can see my hero damage is much higher than anybody in my team, and GPM was close to AM which is very good.
Then in this one http://dotabuff.com/matches/425741824, even though I played better than an average Tinker, my performance was not good, and we lost. My KDA wasn't high, only decent. Yet I got 1281 DSR.
This one http://dotabuff.com/matches/423555508 was a convincing win, but I didn't play godlike. My DSR was 1347.
Doesn't make sense to me.
And why is it ignoring the ranked games?
sano thats cause the system noticed u had no boots on weaver
NO BOOTS STRAT IS LEGIT
I am working on something that will answer most of you criticisms :) I think it will be finally something that will really be "it".
I am sorry that the scores seem unstable - it's still a work in progress and a beta - I use your feedback to improve the whole method. So your criticism is welcome - I read a lot of what you write and try to incorporate as much interesting ideas as I can.
Everyone has bad games, even Funnik, Dendi etc. it's just that they have less of them and in the long run their DSR would be higher. Right now you make conclusions on limited data - 2,3,5 matches while this gets stable at around 20-30. You might have been just unlucky and got a bad game streak. Again, this should be resolved once I optimize the performance a bit.
As for how this currently works:
Right now this version answers two basic questions:
How hard was this game?
How well did this player play given those circumstances?
Then it converts the answer into a score.
So while you may think you've played well, the algorithm looked at your opponent stats and saw that they didn't seem to play top notch dota - that's why you got lower score.
You might have had negative kill:death ratio but the algo saw that opponents played really well ( because they started tryharding, were 5 man stack, all of them picked their best herores, hated you because you insulted them in all-chat so they started playing more intensely etc. ) and the algorithm tried to compensate for this and give you higher score.
Still, tomorrow or in 2 days there will be a new version that I think everyone will find useful, regardless of Valve's MMR :)
"So while you may think you've played well, the algorithm looked at your opponent stats and saw that they didn't seem to play top notch dota - that's why you got lower score."
It gives me >2000 points for a one sided stomp game which I ended with 4-1 score...
How do you get 10 matches to show up?
I went from 514 to 109? really? some games are correct 749... while some where obviously teamrolls, my lich pick with 109,
I got 84 with this match: http://dotabuff.com/matches/423645937
@catman
Its not JUST your performance. Its your performance vs "expected" performance. Which essentially is comparing both lineups to determine who should win, and this will make stomps in games where you should be winning give less points, and losing in matches which you should win less points. Add in performance, and what you get is that because the game was unbalanced in your favor, to get a higher score you need to do a lot better. A 3.1 KDA might be decent in a game which is balanced or unbalanced against you, but if its in your favor, then its not a very good score.
he explains this above
>20:3:10 with PA - seems like a good KDA for this low rating. Well apparently, his opponents weren't that good in that game and he should have probably went 20:1:10 or even 20:0:10 for 1900 rating. Maybe he stacked with worse players or simply faced an opponent(s) with a bad day - because the enemies didn't seem to play top-level dota.
@Terrible, I got 2091 points for this game: http://dotabuff.com/matches/423210041
I think the current algorithm doesn't work properly. Furthermore, I'm playing against people on the first live game pages and my matches get rated between 600 and 1200 points.
On the other side, I raped the #5 player in the list (Ming明明) in two consecutive games (although I lost the second one due to feeding team mates :) ), I think he is pretty overrated.
i had a 10/2/18 game with disruptor, -600 dsr (negative!). this algorithm is horrible. gives supports higher ratings with lower kda's